Citadel: cloud-first HPC, TPUs
What happened
- At Google Cloud Next, Citadel Securities described moving quant research toward a cloud-first HPC model using accelerators like TPUs. - The session highlighted TPUs delivering faster insights and more agile scaling for research workloads. - This indicates peers are shifting parts of quant compute to cloud accelerators for flexible model-driven research (x.com).
Why it matters
Citadel Securities said at Google Cloud Next this week that it is pushing quantitative research toward a hybrid, cloud-first high performance computing setup that uses Google’s Tensor Processing Units. (googlecloudevents.com) High performance computing is the kind of clustered processing used for jobs too large for a single server, and quantitative research is the model-building work trading firms use to price and trade securities. Citadel and Google described the shift as a move from on-premises systems toward cloud capacity that can be added when researchers need it. (googlecloudevents.com) Google scheduled a second Cloud Next 2026 session with Citadel’s head of research platform infrastructure, Haris Nair, focused on “the new Cloud TPU experience.” The session description said TPUs can accelerate “a vast spectrum of complex workloads,” not just the training of large generative artificial intelligence models. (googlecloudevents.com) Google also used Citadel as a customer example in its April 22 Cloud Next announcements. In a company roundup published that day, Google said Citadel built a cloud-based research environment that runs artificial intelligence workloads “up to four times faster” at “30% lower costs,” and that work that once took days now runs in minutes. (blog.google) That pitch lands as Google tries to widen the market for its custom chips beyond chatbot developers. Google announced new Tensor Processing Units at Cloud Next 2026, and CNBC reported on April 22 that Citadel Securities’ quantitative research software already draws on Google TPUs. (cnbc.com) Citadel and Google had already been working on this buildout before this week’s conference. In a Google Cloud case study published in 2024, the companies said they were building a next-generation quantitative research platform to improve researcher productivity and the price-performance ratio of compute. (cloud.google.com) That earlier case study said Citadel’s researchers build models from vast market-data sets to estimate prices across thousands of securities. It also said the firm employs hundreds of quantitative researchers, which helps explain why faster access to shared compute has become a core infrastructure question rather than a side project. (cloud.google.com) Google tied Citadel’s work to a broader infrastructure push at the conference. Another Cloud Next 2026 session featuring Google Cloud chief Amin Vahdat and Citadel Securities chief executive Peng Zhao said the discussion would cover the impact of TPU-based infrastructure on Citadel and “the future of scalable AI.” (googlecloudevents.com) The immediate takeaway from Las Vegas is narrower than “Wall Street is moving to the cloud” and more concrete than that slogan. Citadel is publicly describing a research stack that keeps some high performance computing in-house, adds cloud capacity on demand, and uses TPUs as a faster option for parts of quantitative research. (googlecloudevents.com)
Key numbers
- (googlecloudevents.com) Google also used Citadel as a customer example in its April 22 Cloud Next announcements.
- In a company roundup published that day, Google said Citadel built a cloud-based research environment that runs artificial intelligence workloads “up to four times faster” at “30% lower costs,” and that work that once took days now runs in minutes.
- Google announced new Tensor Processing Units at Cloud Next 2026, and CNBC reported on April 22 that Citadel Securities’ quantitative research software already draws on Google TPUs.
- In a Google Cloud case study published in 2024, the companies said they were building a next-generation quantitative research platform to improve researcher productivity and the price-performance ratio of compute.
What happens next
- Citadel Securities said at Google Cloud Next this week that it is pushing quantitative research toward a hybrid, cloud-first high performance computing setup that uses Google’s Tensor Processing Units.
- (googlecloudevents.com) Google also used Citadel as a customer example in its April 22 Cloud Next announcements.
- Google announced new Tensor Processing Units at Cloud Next 2026, and CNBC reported on April 22 that Citadel Securities’ quantitative research software already draws on Google TPUs.
Quick answers
What happened in Citadel: cloud-first HPC, TPUs?
At Google Cloud Next, Citadel Securities described moving quant research toward a cloud-first HPC model using accelerators like TPUs. The session highlighted TPUs delivering faster insights and more agile scaling for research workloads. This indicates peers are shifting parts of quant compute to cloud accelerators for flexible model-driven research (x.com).
Why does Citadel: cloud-first HPC, TPUs matter?
Citadel Securities said at Google Cloud Next this week that it is pushing quantitative research toward a hybrid, cloud-first high performance computing setup that uses Google’s Tensor Processing Units. (googlecloudevents.com) High performance computing is the kind of clustered processing used for jobs too large for a single server, and quantitative research is the model-building work trading firms use to price and trade securities. Citadel and Google described the shift as a move from on-premises systems toward cloud capacity that can be added when researchers need it. (googlecloudevents.com) Google scheduled a second Cloud Next 2026 session with Citadel’s head of research platform infrastructure, Haris Nair, focused on “the new Cloud TPU experience.” The session description said TPUs can accelerate “a vast spectrum of complex workloads,” not just the training of large generative artificial intelligence models. (googlecloudevents.com) Google also used Citadel as a customer example in its April 22 Cloud Next announcements. In a company roundup published that day, Google said Citadel built a cloud-based research environment that runs artificial intelligence workloads “up to four times faster” at “30% lower costs,” and that work that once took days now runs in minutes. (blog.google) That pitch lands as Google tries to widen the market for its custom chips beyond chatbot developers. Google announced new Tensor Processing Units at Cloud Next 2026, and CNBC reported on April 22 that Citadel Securities’ quantitative research software already draws on Google TPUs. (cnbc.com) Citadel and Google had already been working on this buildout before this week’s conference. In a Google Cloud case study published in 2024, the companies said they were building a next-generation quantitative research platform to improve researcher productivity and the price-performance ratio of compute. (cloud.google.com) That earlier case study said Citadel’s researchers build models from vast market-data sets to estimate prices across thousands of securities. It also said the firm employs hundreds of quantitative researchers, which helps explain why faster access to shared compute has become a core infrastructure question rather than a side project. (cloud.google.com) Google tied Citadel’s work to a broader infrastructure push at the conference. Another Cloud Next 2026 session featuring Google Cloud chief Amin Vahdat and Citadel Securities chief executive Peng Zhao said the discussion would cover the impact of TPU-based infrastructure on Citadel and “the future of scalable AI.” (googlecloudevents.com) The immediate takeaway from Las Vegas is narrower than “Wall Street is moving to the cloud” and more concrete than that slogan. Citadel is publicly describing a research stack that keeps some high performance computing in-house, adds cloud capacity on demand, and uses TPUs as a faster option for parts of quantitative research. (googlecloudevents.com)
